{"id":"W2807926734","doi":"10.1145/3209635.3209655","title":"Exam Wrappers","year":2018,"lang":"en","type":"article","venue":"","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; University of Calgary","funders":"","keywords":"Medical education; Computer science; Drop out; Test (biology); Mathematics education; Psychology; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001604826,0.00002230886,0.00002539083,0.00002272518,0.0004375797,0.0000484253,0.0001224377,0.00002486608,0.009095979],"category_scores_gemma":[0.0009659326,0.00001986459,0.0000129296,0.00008618544,0.0001993032,0.0002837479,0.00001258375,0.00003822023,0.002107348],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002330631,"about_ca_system_score_gemma":0.00006945153,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003137612,"about_ca_topic_score_gemma":0.002740377,"domain_scores_codex":[0.9993299,0.0001978085,0.0000485706,0.00007349495,0.0002501227,0.0001000532],"domain_scores_gemma":[0.9996479,0.0001391467,0.00002887257,0.00007497038,0.0000657696,0.00004332771],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000074452,0.00004463518,0.01299448,0.000001102485,0.000014554,5.239635e-7,0.04884736,7.62766e-7,0.0002841383,0.7209439,0.1622227,0.05463845],"study_design_scores_gemma":[0.00004450015,0.00002119189,0.003091101,0.00000126761,0.00000331077,9.781964e-8,0.0051266,0.00002302857,0.0001447874,0.001664988,0.9898383,0.00004084644],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.07344648,0.000007846079,0.0003483623,0.01926951,0.0003150418,0.0000450386,7.209739e-8,0.0001225195,0.9064451],"genre_scores_gemma":[0.9508851,0.00000641648,0.003163654,0.001004202,0.0004368284,0.000001532358,1.712357e-7,0.000002329572,0.04449974],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8774387,"threshold_uncertainty_score":0.9986696,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2508113948073107,"score_gpt":0.5182574327298312,"score_spread":0.2674460379225205,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}